Validation of China’s 2025 high-risk criteria in RT-PCR-confirmed chikungunya inpatients: a single-centre retrospective cohort

Abstract

Background

China’s 2025 clinical guideline designates age ≥ 65 years, pregnancy, and chronic comorbidity as patient categories requiring prioritized monitoring; quantitative validation in inpatients is lacking.

Methods

We conducted a single‑centre retrospective cohort including 131 consecutive RT‑PCR-confirmed patients admitted during an imported outbreak (1 August−30 September 2025). Patients meeting any high‑risk criterion were classified as high‑risk. The primary endpoint was length of stay (LOS). Secondary endpoints were prolonged hospitalisation (≥ 7 days), prolonged fever (≥ 3 days), and an acute laboratory‑injury composite at admission defined as at least one abnormality among seven routine tests. LOS was modelled using a Gamma generalised linear model (log link). Binary endpoints used robust Poisson regression adjusted for prespecified covariates. For the composite endpoint, available‑case analysis provided the primary estimate (missing D‑dimer values were treated as normal). Multiple imputation (MI; m = 10) and inverse‑probability weighting (IPW) analyses were prespecified sensitivity checks.

Results

Fifty‑three of 131 patients (40.5%) were high‑risk. High‑risk status was associated with longer LOS (time ratio 1.15; adjusted mean + 0.69 days; 95% CI 0.06–0.28). Under MI, high‑risk patients had a higher probability of acute laboratory injury at admission (aRR 1.37; 95% CI 0.87–2.17; P = 0.174). Sensitivity analyses yielded attenuated and method‑dependent estimates; the IPW model gave an aRR of 1.19 (95% CI 0.88–1.61; P = 0.253), indicating sensitivity to missing‑data assumptions. No patients required ICU admission or died (0/131).

Conclusion

China’s 2025 high‑risk criteria identify chikungunya inpatients with longer LOS, supporting targeted early monitoring and resource planning. The admission laboratory‑injury signal was sensitive to missing‑data modelling and should be interpreted cautiously pending external validation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-025-12421-0.

Introduction

Chikungunya fever is an acute arboviral disease transmitted by Aedes mosquitoes that has caused recurrent international outbreaks with expanding geographic range and substantial morbidity, particularly among older adults and those with comorbidities [1, 2]. With increasing travel‑related introductions, several provinces in southern China face seasonal receptivity to local transmission [3–5]. These three high‑risk categories—advanced age, pregnancy, and chronic comorbidity—reflect well‑established determinants of severe arboviral disease. International cohorts have consistently shown that adults ≥ 65 years experience more prolonged arthralgia, higher complication rates, and increased healthcare utilization during chikungunya episodes [6, 7]. Similarly, systematic reviews demonstrate that chronic cardiovascular, metabolic, or respiratory conditions amplify disease severity and delay recovery [2, 8], while pregnant patients face elevated risks of vertical transmission, preterm delivery, and maternal complications [9, 10]. Synthesizing this evidence, China’s National Disease Control Administration issued interim guidance in early 2025 that designated age ≥ 65 years, pregnancy, and chronic comorbidity as high‑risk categories for severe disease requiring prioritized identification and resource allocation [11]. While the guidance targets severe outcomes, it does not specify quantifiable early indicators (e.g., length of stay, laboratory injury) for resource planning in hospitalised patients. However, quantitative validation of these criteria for severe disease among hospitalised patients has not been reported.

For arboviral infections, “high‑risk” stratification typically targets vulnerability to prolonged hospitalisation and organ injury rather than risk of initial infection or ICU admission. Hospital length of stay (LOS) is widely used as a pragmatic marker of acute‑phase burden and resource utilisation in infectious diseases [12–14], capturing both biological severity and operational constraints. These constraints include hospital-specific factors such as bed availability, isolation protocols, weekend discharge policies, and administrative processes that may prolong stay independent of clinical status. In hospitalised chikungunya patients, severe disease risk often manifests as prolonged LOS and early laboratory derangements reflecting endothelial activation and marrow suppression [15–17]. Whether the guideline‑defined high‑risk criteria identify inpatients who experience longer LOS and a higher frequency of these early derangements is a clinically relevant and policy‑salient question, particularly for surge planning during clusters of imported or locally amplified cases. Because isolated laboratory perturbations are common in mild viral illnesses, we hypothesised that a weighted composite emphasizing vascular‑integrity markers would more specifically capture high‑risk chikungunya.

We therefore analysed 131 consecutive RT‑PCR-confirmed inpatients managed during an imported chikungunya outbreak in Fujian in 2025 to validate the predictive value of the national guideline’s high‑risk definition for greater acute‑phase burden. Our primary objective was to compare LOS between high‑risk and non‑high‑risk patients while adjusting for baseline differences; secondary objectives were to assess prolonged hospitalisation and fever, and to quantify an acute laboratory‑injury composite at admission. LOS was selected as the primary endpoint because it integrates biological severity with operational healthcare demands (bed availability, isolation protocols) that dominate outbreak resource planning, whereas ICU admission and mortality were anticipated to be rare events statistically underpowered for validating risk criteria in moderate‑burden chikungunya. We hypothesised that patients meeting any high‑risk criterion would have longer LOS and a higher probability of early laboratory abnormalities, thereby supporting targeted monitoring and near‑term bed‑capacity planning in similar settings. This is a validation study of a risk‑stratification tool, not a predictive inference study; our aim is predictive accuracy rather than estimating a predictive effect.

Methods

Study design and setting

We performed a single‑centre retrospective cohort at the First Hospital of Quanzhou. The study complied with the Declaration of Helsinki and ICMJE recommendations and was approved by the institutional ethics committee ([2025] K310; 9 October 2025), which waived informed consent due to the anonymised retrospective design.

Participants

From 1 August to 30 September 2025, 157 febrile patients with suspected chikungunya presented to the fever clinic; 135 were RT‑PCR-confirmed. Suspected cases were defined per the 2025 Chinese CDC technical guidelines as patients with acute fever (≥ 38 °C) and new-onset polyarthralgia, plus either (i) travel to/return from a chikungunya-endemic country within the previous 12 days, or (ii) documented epidemiological linkage to a confirmed case. We excluded one with an alternative acute febrile illness, one with hospital stay < 24 h, and two with missing key variables, leaving 131 inpatients for analysis (Fig. 1). Two clinicians independently reviewed admission records and investigations, with discrepancies resolved by consensus.

Fig. 1.

Flowchart of patient inclusion and exclusion

Exposure and covariates

Patients meeting any high‑risk criterion (age ≥ 65 years, pregnancy, or chronic comorbidity) were classified as high‑risk; all others were low‑risk. Chronic comorbidity was defined as physician‑diagnosed conditions of ≥ 3 months’ duration encompassing six established risk domains, specifically cardiovascular disease (including hypertension, coronary heart disease, cerebrovascular disease, and heart failure), diabetes mellitus, chronic respiratory disease (including COPD, asthma, and interstitial lung disease), chronic kidney disease of any stage, malignancy either active or within 5 years of treatment, and immunosuppressive conditions such as HIV infection, organ transplantation, or systemic immunosuppressive therapy. Isolated dermatological conditions were explicitly excluded, as they are not established risk modifiers for arboviral disease. Two reviewers independently extracted comorbidity data; patients could have multiple conditions, so Table 1 reports condition‑level (person‑occurrence) prevalence. The full breakdown of specific conditions within each domain is provided in Table S8. Given that this broad definition may capture baseline frailty rather than infection‑specific risk, we adjusted for chronic organ disease and performed sensitivity analyses excluding non‑immune‑relevant comorbidities (Tables S9 and S10). Importantly, chronic comorbidity alone did not trigger the laboratory‑injury composite; only deviations from patient‑specific baselines contributed. We then used a directed acyclic graph (DAG) to map potential confounders and inform a parsimonious adjustment set (Figure S1), including baseline demographics (sex, onset‑to‑admission interval, BMI), pre‑existing conditions (chronic organ disease), and disease status at presentation (admission temperature, white‑cell count, C‑reactive protein, rash). While some markers may lie on a predictive pathway, we included them to control for severity at admission, thereby isolating the incremental predictive value of the high‑risk criteria. A sensitivity analysis excluding these markers is presented in Table S9.

Table 1.

Baseline characteristics and first laboratory values

Variable	Low-risk group (n = 78)	High-risk group (n = 53)	Statistic	P value
Demographics & vital signs
Age, years	35.7 ± 18.2	62.7 ± 18.6	−9.89†	< 0.001
Male sex, n (%)	35 (44.9)	34 (64.2)	4.75‡	0.029
Onset-to-admission time, h	1.0 (0.5, 2.0)	1.0 (0.2, 2.0)	−2.01§	0.044
Admission temperature, °C	36.8 (36.5, 37.6)	36.9 (36.6, 37.2)	−0.04§	0.97
Systolic BP, mmHg	126 ± 18	136 ± 18	−4.20†	< 0.001
Diastolic BP, mmHg	82 ± 13	81 ± 13	−0.53†	0.60
Heart rate, bpm	91 ± 14	85 ± 11	1.86†	0.066
Complete blood count
White blood cell count on admission, ×10⁹/L	4.64 (3.64, 6.04)	5.48 (4.34, 6.82)	−2.12§	0.034
Neutrophil count, ×10⁹/L	2.92 (1.94, 4.22)	3.81 (2.89, 5.26)	−2.99§	0.002
Lymphocyte count, ×10⁹/L	1.16 (0.84, 1.42)	0.87 (0.67, 1.15)	−3.29§	< 0.001
Monocyte count, ×10⁹/L	0.50 (0.40, 0.69)	0.57 (0.41, 0.70)	−1.23§	0.22
Platelet count, ×10⁹/L	216 ± 41	196 ± 57	3.95†	< 0.001
Haemoglobin, g/L	135 ± 14	130 ± 16	0.77†	0.44
Serum biochemistry
Alanine aminotransferase (ALT), U/L	18.0 (11.5, 31.0)	19.5 (14.0, 36.0)	−1.07§	0.28
Aspartate aminotransferase (AST), U/L	25.0 (19.5, 31.0)	26.0 (21.0, 36.0)	−1.48§	0.14
Total bilirubin, µmol/L	10.1 (7.1, 13.2)	12.7 (8.9, 15.1)	−2.37§	0.018
Creatinine, µmol/L	67.7 ± 17.8	80.1 ± 19.0	−3.50†	< 0.001
Sodium, mmol/L	138.7 ± 2.4	136.8 ± 2.5	3.36†	< 0.001
Potassium, mmol/L	3.81 ± 0.41	3.80 ± 0.38	0.24†	0.81
Albumin, g/L	42.4 ± 2.8	39.4 ± 4.0	4.05†	< 0.001
C-reactive protein, mg/L	7.2 (2.8, 17.0)	19.2 (9.7, 31.1)	−4.68§	< 0.001
Coagulation
D-dimer, mg/L	0.42 (0.26, 0.56)	0.45 (0.31, 0.91)	−2.34§	0.019
Prothrombin time, s	11.6 ± 0.9	11.3 ± 1.0	1.84†	0.069
Comorbidities
Hypertension, n (%)	0 (0)	26 (49.1)	—‖	< 0.001
Diabetes mellitus, n (%)	0 (0)	7 (13.2)	—‖	0.003
Coronary heart disease, n (%)	0 (0)	2 (3.8)	—‖	0.11
Other chronic disease,¹ n (%)	0 (0)	8 (15.1)	—‖	0.002

BP, blood pressure; bpm, beats per minute; IQR, inter‑quartile range. Data are mean ± SD, median (Q1, Q3), or n (%). †Student’s t‑test; §Mann-Whitney U; ‡χ²; ‖Fisher’s exact (two‑sided). D‑dimer missing in 15/131 (12 low‑risk; 3 high‑risk); other listed variables had no admission‑time missingness

For the single HIV‑infected patient, baseline thrombocytopenia was documented in outpatient records prior to admission

¹These 8 person‑occurrences comprised: chronic obstructive pulmonary disease (n = 2), rheumatoid arthritis (n = 1), epilepsy (n = 1), HIV infection (n = 1), prior cerebral infarction (n = 1), gastric malignancy (n = 1), and breast cancer (n = 1). No patients with isolated skin diseases such as atopic dermatitis were included. Because patients could have multiple comorbidities, the sum exceeds the number of patients with chronic disease (43)

Outcome

The primary endpoint was LOS, defined as days from admission to discharge. Secondary outcomes included prolonged hospitalisation, defined a priori as LOS ≥ 7 days; prolonged fever, defined as fever duration ≥ 3 days; and an acute laboratory‑injury composite at admission defined as the presence of at least one abnormality among seven routine tests (platelets, ALT, creatinine, sodium, potassium, albumin, and D‑dimer) selected based on documented chikungunya pathogenesis (endothelial activation, marrow suppression, metabolic stress) [18, 19]; and safety outcomes (ICU admission and all‑cause mortality). Total bilirubin was excluded from the composite due to common non‑inflammatory confounders and is reported descriptively at baseline only. Reference intervals and sex-specific ALT thresholds (Table S1) were adopted from the normal ranges quoted in Medical Biochemistry, 5th ed.; the clinical chemistry analyses were performed at Quanzhou First Hospital, 2025.

The ≥ 7‑day threshold was prespecified based on: (1) institutional outbreak‑response protocols requiring isolation‑bed re‑authorisation after one week; (2) arboviral literature where 7 days marks the natural inflection point for clinical course resolution and typical duration of resource‑intensive stays, including ICU care [20, 21]; and (3) operational relevance for bed‑capacity surge planning. This definition was established before data lock and was not derived from the observed LOS distribution.

We selected these seven markers because they capture distinct acute pathophysiological axes directly implicated in chikungunya pathogenesis —endothelial activation/capillary leak (D‑dimer, albumin, sodium), bone‑marrow suppression (platelets), and metabolic/organ stress (ALT, creatinine, potassium) [22, 23]. To ensure the composite reflected acute infection‑related injury rather than baseline frailty, we prespecified that abnormal values in patients with chronic organ disease must represent new decompensation beyond documented baseline (e.g., creatinine rise > 30% from known baseline, or new‑onset thrombocytopenia). The single HIV‑infected patient with baseline thrombocytopenia was excluded from this sensitivity analysis, and all models adjusted for chronic organ disease, yielding consistent estimates (Table S9).

Missing‑data handling for the composite

D-dimer was missing in 15 of 131 patients (11.5%; 12 low-risk, 3 high-risk); other composite markers had no admission-time missingness. We pre‑specified two composite definitions: (1) Primary: clinical‑weight model assigning higher scores to vascular‑integrity markers (thrombocytopenia = 3, elevated ALT/creatinine/hypoalbuminaemia/D‑dimer = 2, hyponatraemia/hypokalaemia = 1), with a positive result defined as total score ≥ 2; the specific weighting scheme was: thrombocytopenia = 3 points; elevated ALT (> 40 U/L male, > 35 U/L female), creatinine > ULN, hypoalbuminaemia (< 35 g/L), or D‑dimer > 0.5 mg/L = 2 points each; hyponatraemia (< 135 or > 145 mmol/L) or dyskalaemia (< 3.5 or > 5.5 mmol/L) = 1 point each; total score ≥ 2 defined as composite‑positive. (2) Secondary: equal‑weight model counting any ≥ 1 abnormality among seven markers. The clinical‑weight model was chosen a priori based on published evidence that platelet suppression and capillary leak drive arboviral severity. Available‑case analysis was used for missing D‑dimer (n = 15), with multiple imputation and inverse‑probability weighting as sensitivity analyses. Imputation included all seven laboratory markers plus age, sex, high-risk status, chronic organ disease, admission temperature, WBC, CRP, LOS and prolonged-hospitalisation indicator. Inverse-probability weighting used a logistic model for D-dimer availability including age, high-risk status, CRP and platelet count. Detailed diagnostics are provided in Supplementary Methods (Supplementary Materials, page 13). Weights were computed as w = 1/π̂_i, where π̂_i denotes the predicted probability of D‑dimer availability; stabilisation applied the factor n/∑w, followed by truncation at the 99th percentile to mitigate extreme values. Stabilised weights were truncated at the 99th percentile (median 1.02, IQR 0.87–1.19; no weight > 10). To address detection-frequency bias we also restricted analyses to patients with all seven tests on day 0–1 (complete-panel, n = 116) and further adjusted for the number of admission-day tests.

Statistical analysis

Descriptive statistics are presented as mean ± SD or median (IQR) for continuous variables and as counts with percentages for categorical variables, with group comparisons conducted using t tests or Mann-Whitney U tests for continuous data and χ² or Fisher’s exact tests for categorical data as appropriate. Because LOS was right‑skewed (Shapiro-Wilk P < 0.001), we modelled it with a Gamma generalised linear model using a log link to estimate time ratios and derived adjusted absolute mean differences in days via smearing of marginal means. Binary endpoints were analysed using robust Poisson regression to estimate aRRs with 95% confidence intervals. The prespecified covariate set for all multivariable models included age, sex, chronic organ disease, admission temperature, admission white‑cell count, and onset‑to‑admission interval. Sensitivity analyses included negative‑binomial models for thresholded composite counts (≥ 1 and ≥ 2 abnormalities), Firth correction for rare events, MI by chained equations with m = 10 imputations with convergence diagnostics provided in the Additional file, IPW for D‑dimer missingness using weights from a logistic model including age, high‑risk status, CRP, and platelet count, truncated at the 99th percentile, with weight distributions shown in the Additional file (Table S7).

We also tested whether the main findings were robust to excluding acute‑phase markers (WBC, CRP, rash) that could be potential mediators, which is relevant for predictive validation. E‑values were computed for the LOS association estimates to assess robustness to unmeasured confounding; E‑values were not calculated for the composite endpoint given its sensitivity to missing‑data handling. PCA with z-scored laboratory variables (KMO = 0.53; Bartlett’s test p < 0.001) retained three components with eigenvalue > 1; loadings and variance explained are reported in Table S2. PC1 (24.2% of variance) summarized an inflammation-organ-injury axis. For multiplicity control, α = 0.05 was set for the primary outcome, while seven exploratory single-marker comparisons were adjusted using Benjamini-Hochberg FDR with q < 0.05 considered statistically significant. P values are reported to three decimals (< 0.001 where applicable), and association estimates with confidence intervals are reported to two decimals. Analyses were performed in SPSS 27.0 and R 4.3.1 with packages sandwich, logistf, mice 3.16, ggplot2, emmeans and dagitty. Reporting adheres to STROBE; the completed checklist is provided in the Supplementary Information.

Results

Study flow and baseline characteristics

Of 157 screened patients, 131 were enrolled (83.4%). By definition, the high‑risk group (n = 53, 40.5%) was older with more comorbidity, and also showed worse inflammatory and nutritional profiles at admission, with lower platelets and albumin and higher CRP and creatinine compared with the low‑risk group (Table 1).

Primary outcome: length of stay

Median LOS was 6 days (IQR 5–7) in high‑risk patients versus 5 days (IQR 3–7) in low‑risk patients (P < 0.001), and LOS distribution was right‑skewed (Shapiro-Wilk P < 0.001). The Gamma model estimated a time ratio of 1.15 (95% CI 1.02–1.31), corresponding to an adjusted mean prolongation of 0.69 days (β = 0.141; 95% CI 0.015–0.267; P = 0.028) (Figure S2).

Secondary outcomes

High‑risk patients were four times as likely to experience prolonged hospitalisation (≥ 7 days) (aRR 4.00, 95% CI 1.29–12.39, P = 0.016), whereas prolonged fever (≥ 3 days) showed no between‑group difference (aRR 0.60, 95% CI 0.10–3.87, P = 0.586). No patients required ICU admission or died (Table 2). In a prespecified sensitivity analysis using a ≥ 6‑day threshold, high‑risk patients remained at increased risk (aRR 1.54, 95% CI 1.05–2.26, P = 0.027); as expected, the magnitude was attenuated, supporting the specificity of the day‑7 operational cut‑point for capturing resource‑intensive stays. Laboratory‑injury patterns told a more nuanced story. When we scored admission labs using clinical weights that prioritized vascular‑integrity markers (platelets, albumin, D‑dimer), high‑risk patients surpassed the ≥ 2‑abnormality threshold far more often than low‑risk patients (62.3% vs. 39.7%; aRR 1.49, 95% CI 1.06–2.08, P = 0.021). By contrast, a simple count of any one abnormal test (≥ 1/7) yielded a weaker, non‑significant gradient (81.1% vs. 64.1%; aRR 1.27, 95% CI 0.81−2.00, P = 0.298), confirming that cumulative injury burden, not isolated perturbations, marks high‑risk disease.

Table 2.

Individual laboratory abnormalities and secondary outcomes

Outcome	Low-risk (n = 78)	High-risk (n = 53)	aRR (95% CI)	P value
Secondary clinical outcomes
Hospital stay ≥ 7 days (primary)	5 (6.4%)	15 (28.3%)	4.00 (1.29–12.39)	0.016
Hospital stay ≥ 6 days (sensitivity)	31 (39.7%)	32 (60.4%)	1.54 (1.05–2.26)	0.027
Fever ≥ 3 days	5 (6.4%)	2 (3.8%)	0.60 (0.10–3.87)	0.586
ICU admission or death	0 (0%)	0 (0%)	-	-
Acute laboratory injury composite
Clinical-weight model, ≥ 2 abnormalities	31 (39.7%)	33 (62.3%)	1.49 (1.06–2.08)	0.021
Equal-weight model, ≥ 1 abnormality	50 (64.1%)	43 (81.1%)	1.27 (0.81−2.00)	0.298
Single abnormalities
ALT > ULN	10 (12.8%)	8 (15.1%)	1.52 (0.58–4.02)	0.394
Serum creatinine > ULN	11 (14.1%)	6 (11.3%)	0.97 (0.32–2.88)	0.949
Hyponatraemia	3 (3.8%)	11 (20.8%)	4.93 (1.09–22.26)	0.010
Hypokalaemia	14 (17.9%)	9 (17.0%)	1.04 (0.41–2.65)	0.937

aRR: adjusted risk ratio from robust Poisson regression adjusted for age, sex, chronic organ disease, admission temperature, white-cell count, and onset‑to‑admission interval

Clinical‑weight model assigned weights: thrombocytopenia = 3, elevated ALT/creatinine/hypoalbuminaemia/D‑dimer = 2, hyponatraemia/hypokalaemia = 1; composite positive if total score ≥ 2

Equal‑weight model scored positive if ≥ 1 of 7 abnormalities present

D‑dimer missing in 15/131 patients; treated as normal (available‑case). All other tests were complete

Sensitivity analysis: Prolonged hospitalisation (≥ 6 days) - Low‑risk 18/78 (23.1%), High‑risk 25/53 (47.2%), aRR 1.54 (95% CI 1.05–2.26), P = 0.027

Bold row indicates the primary composite endpoint

Zooming in on individual markers, only hyponatraemia survived Benjamini-Hochberg FDR correction (20.8% vs. 3.8%; raw P = 0.010, q = 0.024); all other single abnormalities were non‑significant (Table 2, Table S3).

Sensitivity analyses

For the composite endpoint, available-case and complete-case estimates were attenuated relative to the MI primary analysis and had wider intervals.The IPW model yielded an aRR of 1.19 (95% CI 0.88–1.61; P = 0.253), with the 95% CI spanning 1 and the point estimate differing in direction from the MI estimate, indicating sensitivity to the assumed missing-data mechanism and limited covariate overlap (Table S4). To address potential detection-frequency bias, the complete-panel restriction (n = 116) produced an aRR of 0.72 (95% CI 0.46–1.12; P = 0.146), and the model additionally adjusting for the number of admission-day tests gave an aRR of 0.75 (95% CI 0.49–1.13; P = 0.171). For LOS, exclusion of pregnancy and alternative count-link specifications did not materially change effect direction or magnitude.

Subgroup heterogeneity

Exploratory mutually-exclusive strata considering a single criterion only showed no significant adjusted mean LOS differences versus non-high-risk patients: elderly-only−1.14 days (95% CI−3.82 to + 4.38, P = 0.59), pregnancy-only−0.93 days (−3.78 to + 5.16, P = 0.68; underpowered, n = 1), and chronic-disease-only−0.23 days (−3.42 to + 6.53, P = 0.93). Interaction terms were non-significant (elderly × chronic disease P = 0.75; pregnancy × chronic disease P = 0.59) (Table S5).

Inflammation-organ‑injury axis

PC1 had an eigenvalue of 1.70 explaining 24.3% of the variance, driven by D‑dimer (negative loading), platelets, and albumin, consistent with a capillary‑leak-bone‑marrow‑suppression axis. High‑risk patients had higher PC1 scores than low‑risk patients (median 0.41 vs−0.47; P < 0.001) (Table S2).

Discussion

During a 2025 imported chikungunya outbreak in our single-centre inpatient cohort, China’s high-risk criteria (age ≥ 65 years, pregnancy, or chronic comorbidity) identified patients with marginally longer admissions. High-risk status was associated with an adjusted mean prolongation of 0.69 days (95% CI 0.06–1.32, P = 0.028), representing a 15% increase in LOS. This magnitude aligns with international reports for dengue and other arboviral infections, where guideline-defined vulnerable groups typically experience 1–2 extra bed-days (Table S6) [24–26]. While clinically modest, this translates to meaningful capacity planning: approximately 7 additional bed-days per 10 high-risk admissions. The ≥ 7-day dichotomous endpoint operationalises this as a trigger aligned with our hospital’s isolation-bed policy, where stays exceeding one week require re-authorisation. This pragmatic focus on resource utilization, rather than mortality, is reinforced by the complete absence of ICU admissions or deaths (Table 2). China’s high‑risk criteria therefore predict moderate burden rather than fatal outcomes, validating LOS as a relevant primary endpoint for low‑mortality arboviral outbreaks. Predicting resource burden rather than mortality, this performance profile contrasts sharply with dengue severity criteria, which prioritize dynamic warning signs (abdominal pain, persistent vomiting, hematocrit rise) to predict life‑threatening hemorrhage and shock [27, 28]. While dengue tools excel at ICU triage and mortality reduction, our findings position chikungunya high‑risk criteria as a complementary tool for outbreak surge capacity planning, defining a distinct clinical niche that is more valuable for predicting bed‑days than for mortality risk stratification in low‑fatality settings. This zero‑event rate underscores that these criteria are not designed to predict mortality or ICU admission, but rather to forecast resource utilization in inherently low‑fatality chikungunya outbreaks. Future validation in higher‑mortality contexts is needed to assess generalizability for severe outcome prediction.

Laboratory findings corroborated this gradient. A clinical‑weight composite capturing cumulative vascular‑marrow injury (platelets, albumin, D‑dimer) was significantly more prevalent in high‑risk patients (≥ 2 abnormalities: 62.3% vs. 39.7%; aRR 1.49, P = 0.021). High-risk patients also presented more frequently with hyponatraemia (20.8% vs. 3.8%; q = 0.024) and non-significantly lower platelets and albumin levels, a pattern consistent with endothelial activation, capillary leak, and transient bone-marrow suppression [29–33]. Principal component analysis reinforced this signature: Axis 1 (driven by platelets, albumin, and D-dimer) explained 24% of laboratory variance and scored higher in high-risk patients (median 0.41 vs. -0.47, P < 0.001), further supporting an inflammation-vascular integrity signature described in arboviral fevers.

The prespecified ≥ 7-day threshold proved more specific for high-resource utilisation than a ≥ 6-day cut-point, which diluted the effect size (aRR 1.54 vs. 4.00) by capturing uncomplicated recovery trajectories. This supports our a priori operational definition. To address potential confounding, we performed a sensitivity analysis excluding WBC, CRP, and rash—markers that could represent disease severity rather than baseline risk. The association not only persisted but strengthened (adjusted time ratio 1.18, 95% CI 1.04–1.32, P = 0.008), demonstrating robustness. DAGs transparently guided confounder selection for risk adjustment, a method appropriate for predictive modelling [34]. These robust analytical foundations support translation into clinical protocols.

Integrating these criteria into practice requires a tiered response protocol. At triage, electronic health records should flag patients meeting any high‑risk criterion, triggering enhanced monitoring: (1) Age ≥ 65 years: mandatory 8‑hourly temperature and fluid‑balance charting, plus baseline ECG and renal function within 24 h; (2) Pregnancy: immediate obstetric review and fetal monitoring every 48 h; (3) Chronic comorbidity: admission D‑dimer, platelet, and sodium panels, with repeat testing at day 3 if the composite score is borderline. A composite score ≥ 2 should activate a “red‑alert” pathway: infectious‑disease consultant review, consideration of isolation‑ward transfer, and protocolized intravenous fluid management targeting urine output > 0.5 mL/kg/h. Given the significant hyponatraemia gradient (20.8% vs. 3.8%; q = 0.024), electrolyte monitoring should be standard for high‑risk admissions. For LOS ≥ 7 days, weekly multidisciplinary review is recommended. Operationally, 7 additional bed‑days per 10 high‑risk admissions should be budgeted. The clinical utility of this framework hinges on the underlying composite endpoint’s methodological robustness, which we examined through multiple sensitivity analyses.

The composite “acute laboratory injury” endpoint proved highly method-sensitive: the simple count (≥ 1 abnormality) yielded unstable estimates across analytical strategies (aRR 1.27 in primary analysis, but ranging 1.19–1.54 in sensitivity analyses, Table S4). In contrast, the clinical-weight ≥ 2 model—prioritising vascular-integrity markers—was robust. Using an available-case approach (primary), 62.3% of high-risk vs. 39.7% of low-risk patients met this threshold (aRR 1.49, 95% CI 1.06–2.08, P = 0.021) (Table S7). Multiple imputation and complete-case analyses produced near-identical point estimates (aRR 1.51 and 1.55, respectively), confirming stability. After restricting to the 116 patients with full seven-test panels and adjusting for test frequency, the association persisted (aRR 1.44, 95% CI 0.98–2.13), arguing against detection bias. These data support that a biologically driven, weighted composite captures prognostically meaningful injury better than simple abnormality counts.

Our suspected-case definition introduced spectrum bias. The 86% RT-PCR confirmation rate (135/157) reflects a highly specific definition aligned with the 2025 Chinese CDC guideline: acute fever with arthralgia plus recent travel linkage. While this optimises diagnostic efficiency during outbreak containment, its high specificity sacrifices sensitivity, potentially missing mild or atypical infections. Consequently, our cohort likely over-represents patients with pronounced manifestations who met testing criteria, biasing our sample toward greater severity. Conversely, we may have missed pauci-symptomatic cases. False-negative PCR is unlikely given our short onset-to-admission interval (median 1 day). Thus, our findings generalise to symptomatic, travel-associated chikungunya requiring inpatient management; prospective studies with broader screening criteria (e.g., fever alone in endemic settings) are needed to validate performance in community-managed cases.

Several limitations merit consideration. First, our single-centre retrospective design inherently limits generalisability and introduces potential selection bias. Patients were drawn from a single tertiary hospital during an imported outbreak, which may not represent the broader epidemiology of chikungunya in different settings (e.g., endemic regions with local transmission, community-managed cases, or diverse healthcare systems). The retrospective nature also constrained our ability to standardise data collection and may have introduced information bias. Second, a notable limitation was the 11.5% missing D-dimer data (n = 15). While IPW was applied, sensitivity analyses revealed marked estimator instability: IPW yielded a substantially attenuated association (aRR 1.19, 95% CI 0.88–1.61) compared to multiple imputation (aRR 1.37) and complete-case analyses (aRR 1.54), suggesting the missing-at-random assumption may be violated if clinicians selectively ordered D-dimer based on unmeasured severity indicators. Consequently, laboratory-injury findings remain sensitive to missing-data assumptions even after IPW and should be considered hypothesis-generating. Third, retrospective comorbidity ascertainment risks misclassification; the guideline’s inclusive definition may capture baseline frailty rather than chikungunya-specific vulnerability. Fourth, our suspected-case definition introduced spectrum bias, over-representing severe cases while missing mild/atypical infections. Fifth, the absence of ICU admissions or deaths limits inference about the criteria’s utility for mortality prediction. Finally, the small pregnancy stratum (n = 2) precluded meaningful subgroup analysis. These limitations underscore the critical need for prospective, multi-centre validation across diverse geographic regions and healthcare settings to confirm the generalizability of our findings.

In conclusion, China’s 2025 high-risk definition is independently associated with marginally longer admissions and early hyponatraemia, providing actionable evidence for early triage and bed-capacity buffering. The composite laboratory-injury signal, while consistent with arboviral pathophysiology, is sensitive to missing-data assumptions and should be considered hypothesis-generating pending multicentre validation. Establishing harmonised admission laboratory panels and agreed minimal clinically important differences for LOS will strengthen the translational utility of high-risk stratification for future chikungunya outbreaks.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

aRR

Adjusted risk ratio

ALT

Alanine aminotransferase

BH‑FDR

Benjamini-Hochberg false‑discovery rate

CI

Confidence interval

DAG

Directed acyclic graph

GLM

Generalised linear model

ICMJE

International Committee of Medical Journal Editors

IPW

Inverse‑probability weighting

IQR

Inter‑quartile range

LOS

Length of stay

MCID

Minimal clinically important difference

MI

Multiple imputation

PCA

Principal‑component analysis

PCR

Polymerase chain reaction

ULN

Upper limit of normal

Author contributions

S.Y. conceptualised the study, analysed data, drafted and revised the manuscript. Z.W. and L.Z. curated data, produced tables/figures, and cross-validated analyses. H.Z. collected clinical variables, adjudicated endpoints, and interpreted imaging. Y.L. and Z.Weng processed laboratory specimens, verified results, and compiled biomarker datasets. X.W. and Z.S. reviewed imaging-histology correlations and critically revised the manuscript. X.Y. acquired funding, oversaw project administration, and guaranteed the final content. All authors participated in patient care and read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (82570707, 82370604), the Major Science and Technology Innovation Project of Fujian Province (2023Y9269), the Natural Science Foundation of Fujian Province (2023J01239), the Medical Innovation Project of Fujian Provincial Health Commission (2015CXA058), and the Climbing Project of the Medical-Education Integration Fund of Fujian Normal University (2025YJRHPD001). The funders had no role in study design, data collection, analysis, interpretation, writing, or the decision to publish.

Data availability

A de‑identified analytic data dictionary and statistical scripts (R/SQL) are available from the corresponding author upon reasonable request. Raw medical records will not be publicly released due to privacy and ethical constraints.

Declarations

Ethics approval and consent to participate

This study protocol was reviewed and approved by the Medical Ethics Committee of the First Hospital of Quanzhou, Fujian Medical University (approval number: [2025] K310; date: 9 October 2025). The study was conducted in accordance with the ethical principles of the Declaration of Helsinki (https://www.wma.net/policies-post/wma-declaration-of-helsinki/). Given the retrospective, anonymized nature of the data analysis, the Ethics Committee granted a waiver of informed consent in compliance with national regulations and institutional policies for retrospective observational studies. All personal identifiers were removed before data analysis to ensure participant confidentiality.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.